# The detailed tour: claims, theorems, and the measurement story *(The bridge page between the introductory [`README`](../README.md) and the modules. Everything here names its theorem. Sector-by-sector reading paths: [`PATHS.md`](PATHS.md). The A1–A7 axiom-level audit: [`specs/reconstruction-status.md`](../specs/reconstruction-status.md).)* ## The reconstruction in four hundred words **What is posited.** One object: a state space of possible ways the world can be, carrying a natural notion of volume and a deterministic law of motion that preserves it. The state space is a projective space of rays together with a small torus of further coordinates. Nothing in it is probabilistic. A preparation is a region of that space; the experimenter does not know which point in the region the world occupies. Probability is read as a fraction of volume, which is the same move that makes statistical mechanics work, applied to a different space. **What is derived, and how generally.** Given the posit, the following are theorems rather than assumptions. The evolution induced on rays is the Schrodinger equation for an arbitrary Hermitian generator, given that the projected flow is in the unitary class, which is itself posited (see below). The probability of an outcome is the volume fraction of the region realising it, and that fraction equals the Born weight, in every finite dimension, for arbitrary unit preparations, with no positivity side condition, and for generalised measurements as well as sharp ones. Observed frequencies over repeated preparations converge to those weights. Measurement is a physical interaction: an explicit volume-preserving dynamics creates a record, the record persists in the specific sense given below, distinct outcomes exclude one another, and the post-measurement state update falls out as a pushforward rather than being postulated. What looks like collapse is relocation of information with storage, not destruction. **What is posited and not derived.** The state space itself. Its existence, dimension, and product structure are assumed, and this document does not pretend otherwise. The typicality reading, that volume fraction is probability, is a hypothesis on the types rather than a consequence. The **unitary class**: that the projected flow is projectively unitary is assumed, not shown. `projectedFlow_eq_unitary_family` takes that as a hypothesis and selects the family, and the generator then follows, so what the corpus proves is "unitary implies Schrodinger", not "the flow is unitary". The assumption is not arbitrary: without it, nonlinear projective evolution combined with tensor structure and local Born statistics permits signalling (Gisin 1990), which no-signalling forbids. A derivation from no-signalling is available if wanted, and until it is done this stands as a posit. The exact-record measurement witnesses require a calibration that is posited. **Persistence of records** is register-freezing by construction, not apparatus memory. In the witness model the interaction carries the register coordinate through unchanged and the readout never inspects it, both of which are theorems rather than readings, so a record survives because nothing in the model can disturb it. A pointer that physically latches and holds would have to write to the register and read from it, and this model does neither. Generic recurrence is also deferred: a measure-preserving flow on a finite measure space returns, by Poincare and quantitatively by Kac, so indefinite persistence does not follow from conservation alone. What would break recurrence is record proliferation, and that tier is not built. A proved trade-off shows that continuity of the dynamics and exactness of records everywhere cannot both hold, so each measurement witness gives up exactly one of continuity, exactness of the Born weights, or exactness at a calibrated point. **Equivariance.** The Bohmian analogue holds and is named. Under the isolated flow the epistemic measure is carried to the epistemic measure at the evolved ray, exactly as Bohmian mechanics carries the squared amplitude to its evolved value, while the typicality measure is left invariant. Both halves are theorems on the concrete arena, with no posit inside either (`csd_equivariance`). Two things it does not say: the epistemic measure and the typicality measure behave differently, one moving and one fixed, so they must not be run together; and preservation does not select the measure. That selection is symmetry, and a deterministic flow provably does not pin it. **Where this sits.** Like Bohmian mechanics, this surrenders ontic factorisability: the underlying state is not a product over subsystems, and that is what buys no-signalling of the marginals without locality of the state. The precision matters, because three different things get called locality here: no-signalling of the marginals, which is what is proved and only conditionally; independence of the Bell parameter, which is denied as a theorem; and joint locality in the local-friendliness sense, which the corpus does not address. Only the first is claimed. Unlike Everett, there is one world; the branching structure appears as regions of a single trajectory space, and outcomes are selected by which region the actual point occupies. Unlike Gleason-style and Busch-style derivations, the general dimension Born result here does not route through effect-Gleason; it is a volume computation. A non-contextuality assumption does remain, in the form of the posited sector structure, and that dependence is common to the whole Gleason-family of derivations rather than specific to this programme. ## The claim, precisely Constraint-Surface Dynamics provides a machine-checked reconstruction of the principal finite-dimensional measurement statistics and update structure on its stated ontic witness: deterministic ontic dynamics give definite context-indexed outcomes without stochastic collapse or branching; Born weights follow from the Fubini-Study/Kähler moment-map geometry and are realised as the measures of explicit ontic fibre outcome cells rather than postulated as an independent probability law; and outcome conditioning reproduces the Lüders update. Basis-selective decoherence is proved for a supplied measurement context. In detail: on the ontic surface `Σ = ℂℙ^{N-1} × T²` with its Liouville measure `μL = μ_FS ⊗ vol` and the projection `π` to rays, the corpus machine-verifies, for every dimension `N`, every unit preparation, and arbitrary Hermitian `H`: - **isolated dynamics** projects to Schrödinger evolution `exp(-itH)` on rays (`projectedFlow_schrodinger_form`, instantiated non-trivially on `manyToOneSchrodingerSetup`); - **the Born rule is a theorem**: `‖⟨eᵢ,ψ⟩‖²` is a Fubini–Study typicality volume, Gleason-free, POVMs included (`fs_born_volume_ratio_N_uncond`, every unit preparation, vanishing weights included, and `povm_born_frequency_volume`), with i.i.d. frequencies converging to it almost surely; - **measurement is a process**: an explicit measure-preserving propagator carries a positive-measure apparatus-ready state into created, persistent, mutually exclusive records whose outcome sectors carry exactly the Born weights (`SwapMeasurementClosure`, `swap_sector_born`); - **the state update is a pushforward theorem**: rank-one Lüders (`swap_luders_born`), and, via the projective join `ℙ(ℂ^{N+N})`, **degenerate** Lüders (`joinWitness_blockLuders` inhabits `BlockLudersObligation`), each through Liouville-preserving dynamics; - **the apparatus basis is not preferred structure**: the full measurement closure holds for every orthonormal basis (`measurement_covariance`); - **one arena carries isolated dynamics plus the complete rank-one measurement reconstruction** (`unifiedArenaClosure`): the isolated flow and the record-creating propagator preserve the *same* Liouville measure, and the isolate → measure → isolate round trip is a theorem (`arena_round_trip`). The degenerate update runs on the companion projective-join witness, and a single capstone bundling rank-one, all bases, and degenerate is delivered in the next bullet. - **one theorem now carries the whole projective-measurement layer** (`projectiveMeasurementCapstone`, `RecordLayer/MeasurementCapstone.lean`): for every Hermitian generator, base point, and unit preparation, rank-one on the unified arena, the six-fact closure in **every** apparatus basis, the **complete degenerate package on one protocol** for every block structure (records, exclusivity, persistence, Liouville, the coarse Born mass, ψ-dependent Lüders, `DegenerateMeasurementClosure`, upgraded 2026-08-03), the smooth `ε`-horn, and its Schrödinger generation. The fields quantify over different witnesses by design (the two-horn framing); new prose cites this theorem and the constituent closures remain as the construction record; - **no-signalling holds dynamically, in every basis** (`reduceA_localLudersOn_mixture`, with the local-measurement-is-block-measurement bridge `toComposite_blockProj` and the join witness supplying the post-states): the Born-weighted mixture of local marginals after a distant measurement equals the marginal before, the distant party's outcome *and basis choice* are invisible. **Mixed preparations** run through the same dynamics: `mixedSwapPrep ρ (outcomeSector i) = Tr(ρ|eᵢ⟩⟨eᵢ|)` (`mixed_swap_sector_born`), and the outcome-conditioned update is delivered too: the post-ensemble is the Bayes-posterior mixture (`mixed_post_bayes`), and at rank one **the record erases the classical ignorance**, follow-up statistics equal the pure Lüders update's (`mixed_luders_followup`); - **the quantum eraser is a process, and records are irreversible**: the mark stroke produces which-path states with flat statistics at every phase (`marked_no_fringe`); the erase stroke on the *coherent* state produces exactly the conditioned fringe states (`erased_amp`, the dynamical post-states are `√2·eraserOut`, dark zero included); and erasing **after** the mark record revives nothing (`sequential_no_revival`), interference is recoverable only before a record exists; - **POVMs and instruments are dynamical too** (`povm_selector_born`, `povm_instrument`, `naimarkInstrumentClosureCanonical`, `RecordLayer/PovmDynamics.lean`): Naimark-dilate and run the *existing* degenerate record protocol on the dilated arena, the ancilla block structure IS `localBlock`, so no new dynamics or sectors are needed. Outcome sectors carry `⟨ψ, Eᵢ ψ⟩` exactly, and the post-states the join witness delivers are the Naimark–Lüders instrument posts `Πᵢ(Vψ)`. Every POVM, via the canonical dilation (the instrument is dilation-relative, a POVM does not determine its instrument); - **measurement dynamics has a smooth horn** (`smoothWitnessClosure`, `RecordLayer/PointerBorn.lean`): on the pointer arena `ℂℙ^{N-1} × T² × ℂℙ^N`, one witness simultaneously carries a measurement protocol whose two-time law is the exponential group property, a propagator **jointly continuous in time and state** (`continuous_pointerRampedEvolve`), Liouville preservation, a positive-measure ready state with no Dirac calibration posit, record creation with the ontic sector selecting the outcome, structural persistence, and the Born sandwich `rⱼ − 2ε ≤ sector ≤ rⱼ + 2(N−1)ε`. At **every** time the propagator satisfies the **Schrödinger equation** with the explicit Hermitian coupling `H_eff` and the rate factor `smoothTransition′(t)` (`rampedU_schrodinger`; window-free since B1b, 2026-08-04), the Hamiltonian-generation statement at the formalisable level. *Precision (2026-08-03, fourth external review): the generation is **fibrewise**, the joint-arena flow's register back-reaction is suppressed by design (`pointerEvolve_base_marginal_unchanged` is the fingerprint; `ι_Vω ≠ d𝓗` on the ε-collars, genuinely Hamiltonian off them), and the original trapezoid weight/ramp ingredients are proved `Continuous`, not `C^∞`, *discharged same day at the ingredient level*: `PointerSmoothProfile.lean` provides `Real.smoothTransition` profiles with the identical plateau interface, a `C^∞` weight lift (`contDiff_smoothArcWeight_lift`), and the Schrödinger equation at **every** time (`smoothRampedU_schrodinger`). The fibrewise (not joint-arena Hamiltonian) character is unchanged; see `PointerGeneration.lean`'s honest-scope block.* - **the de-isolation interface is populated by the dynamics** (2026-08-27, `RecordLayer/ShearDeIsolation.lean`): the pointer of `shearDeIsolationInteraction` is the readout of the constructed propagator, pointwise (`cellPointer_outcomeSector_eq_readout`), and its basins carry the moment-map weights under the ontic ready preparation (`shear_sector_born`, `shearDeIsolation_born`), the state entering only through the preparation measure while readout arcs, basins, and propagator stay context-fixed. The **einselection criterion is a characterisation**: the pointer observables of a supplied interaction are exactly the commuting ones (`pointer_invariant_iff_commute`, with the disturbed contrast `einselection_commutation_contrast` and coherence survival under the flow, `coherence_modulus_preserved`; `Empirical/CSD/PointerCommutation.lean`). The **externality pair** joined the dynamical bundle (`outcome_system_dependent`: before the stroke a system-only change moves the outcome; the after half is recorded as vacuous by architecture). Standing residue, carried not laundered: the interaction is an input, and the propagator's Hamiltonian generation is stated, not formalised (the manifold symplectic API exists since 2026-09-08, `HamiltonianVectorField.lean`; the propagator is not yet stated in it; `ShearWitness` item 1). ## The results, by pillar | Reconstructed pillar | Headline theorem | Module | |---|---|---| | Schrödinger evolution from the sector flow (**from continuity alone** since 2026-09-01) | `projectedFlow_schrodinger_form_of_continuous_flow` (neither the coboundary nor the smoothness datum is posited), `projectedFlow_schrodinger_form` (the original, both staged), `manyToOneSchrodingerSetup_both_pillars` | `LF4/PhaseLift`, `Mathlib/Analysis/Matrix/{ProjectiveLift,StoneC1}`, `LF4/…` | | Born rule as FS typicality volume (all `N`, POVMs, zero weights included) | `fs_born_volume_ratio_N_uncond`, `povm_born_frequency_volume` | `LF4/…` | | The sector flow is Hamiltonian on the sector's own manifold (since 2026-09-09): for every Hermitian `H`, `exp(-itH)` on `ℂℙⁿ` is the integral curve of the Hamiltonian vector field of `-2⟨H⟩` for the Fubini–Study symplectic form, `⟨H⟩` conserved; the torus action is the diagonal case with the Born weights' moment map as its Hamiltonian, unique up to a constant, image the simplex | `schrodingerField_isHamiltonianVectorField`, `isMIntegralCurve_schrodingerUnitary_smul`, `torusField_isHamiltonianVectorField`, `eq_torusHamiltonian_of_nonneg_of_sum`, `range_momentMap` | `Mathlib/Geometry/Manifold/Instances/ProjectiveSpaceSchrodingerFlow`, `…ProjectiveSpaceMomentMap` | | Fubini–Study bridge `π_*μL = μ_FS` | `productSector_hasFubiniStudyPushforward`, `arenaRay_pushforward` | `SigmaLayer/MeasureBridge`, `UnifiedArena` | | Context-fixed measurement partitions (Paper C A7) | `globalBasin_born`, `globalBasin_prob` | `SigmaLayer/GlobalBasin` | | Records created, persistent, exclusive, dynamically | `SwapMeasurementClosure` / `swapMeasurementClosure` | `SigmaLayer/SwapClosure` | | Dynamical Born (outcome-sector measure = Born weight) | `swap_sector_born`, `sector_born_ctx` | `SigmaLayer/SwapClosure`, `RotatedSwap` | | Rank-one Lüders as pushforward | `swap_luders_born` | `SigmaLayer/SwapLuders` | | Degenerate Lüders (the projective join) | `joinWitness_blockLuders`, `join_block_luders`, `joinSwap_measurePreserving` | `SigmaLayer/JoinLuders`, `JoinArena` | | Unitary covariance of measurement | `measurement_covariance` | `SigmaLayer/RotatedSwap` | | One arena, one Liouville measure family (rank-one tier) | `unifiedArenaClosure`, `arena_round_trip` | `SigmaLayer/UnifiedArena` | | Smooth measurement dynamics: Schrödinger-generated, jointly continuous | `rampedU_schrodinger`, `continuous_pointerRampedEvolve` | `SigmaLayer/PointerGeneration`, `PointerProtocol` | | The ε-Born sandwich and the smooth-horn closure | `pointer_born_lower`/`pointer_born_upper`, `smoothWitnessClosure` | `SigmaLayer/PointerBorn` | | The projective-measurement capstone (five fields, one theorem) | `projectiveMeasurementCapstone` | `SigmaLayer/MeasurementCapstone` | | The degenerate one-protocol package; the coarse dynamical Born mass | `degenerateMeasurementClosure`, `join_sector_born` | `SigmaLayer/JoinClosure` | | Dynamical no-signalling, every basis; local = block-degenerate | `reduceA_localLudersOn_mixture`, `toComposite_blockProj` | `SigmaLayer/LocalLudersBasis`, `LocalBlockBridge` | | The eraser as a process; statistical irreversibility of records | `erased_amp`, `erased_rate`, `sequential_no_revival` | `Empirical/CSD/EraserDynamics`, `EraserSequential` | | Mixed preparations: the mixed dynamical Born rule | `mixed_swap_sector_born` | `SigmaLayer/MixedSwap` | | The outcome-conditioned mixed update: Bayes posterior + rank-one collapse of ignorance | `mixed_post_bayes`, `mixed_luders_followup` | `SigmaLayer/MixedLuders` | | The POVM Born rule (selector level); the Naimark–Lüders instrument | `povm_selector_born`, `povm_instrument`, `naimarkInstrumentClosureCanonical` | `SigmaLayer/PovmDynamics` | | `C^∞` transition profiles: smooth weights (universal-cover `C^∞`) and Schrödinger at every time, with the *time-dependent* generator `smoothTransition′(t)·H_eff` | `contDiff_smoothArcWeight_lift`, `smoothRampedU_schrodinger` | `SigmaLayer/PointerSmoothProfile` | | Back-reaction is harmless: a joint lift's record preimages are the SAME SET as the fibrewise witness's | `IsJointLift.outcomeSector_eq`, `jointFlowTransfer` | `SigmaLayer/JointFlowTransfer` | | The third horn: continuous, null (two-point) seam, exact Born (of a free cell-split parameter, the module has no preparation or moment map), at Dirac calibration | `nullSeamClosure`, `nullSeam_born_left` | `SigmaLayer/NullSeamWitness` | | The Kähler triple, pointwise (`g`, `ω`, `J`); at manifold level `ℂℙⁿ` with the Fubini–Study form is a symplectic (2026-09-07) and Kähler (2026-09-10) manifold whose top power is `(4π)ⁿ μ_FS`; and the sector's Liouville measure IS that normalised top power, its unitary flows preserving it | `isFubiniStudyKahler`, `fsForm_isSymplectic`, `fsForm_isKahler`, `fsVolume_eq_smul_fsMeasure`, `unitaryFlowSetup_isKahler_liouville` | `Mathlib/Analysis/InnerProductSpace/KahlerForm`, `Mathlib/Geometry/Manifold/Instances/ProjectiveSpaceFubiniStudySymplectic`, `LF4/SectorManifold` | | Repeatability & sequential statistics | `csd_repeatability`, `csd_sequential_born` | `Empirical/CSD/SequentialMeasurement` | | Mixed states, weights and frequencies | `mixed_ontic_born_weight`, `arena_mixed_born_frequency` | `SigmaLayer/MixedOntic`, `UnifiedArena` | | Entanglement / non-locality / no-signalling | `no_product_partition_realises_singlet`, CGLMP ∀`d`, GHZ ∀`n` | `LF6/…` | | Contextuality, Bell, Tsirelson, uncertainty, thermodynamics TH1–TH4 | see [`EMPIRICAL.md`](../EMPIRICAL.md) and the module headers | `Empirical/…`, `Thermo/…` | | **From the posits to the quantum-information layer** (2026-09-11/12): the density operator of a preparation on `Σ` IS the barycentre of its rays; channels are the environment marginals of `Σ`-flows lifting a unitary (the measurement flow yields the de-isolation channel, unconditionally); the second law, data processing and Landauer on `Σ`; concavity and the Holevo bound with no support hypothesis, the de-isolation channel's capacity = one bit; the three-qubit code on `Σ` end to end (error = a flow's leak into the environment, syndrome = a partition of `Σ`, recovery = one channel) | `preparationDensity_eq_barycenter`, `traceRight_barycenter_flow`, `measurementFlow_traceRight_barycenter_unitSection`, `vonNeumannEntropy_le_pinching_flow`, `landauer_flow`, `deisolationChannel_holevoCapacity`, `registerFlow_recovery` | `LF2/PreparationBarycenter`, `LF2/FlowChannel`, `LF6/MeasurementFlowChannel`, `Thermo/SigmaSecondLaw`, `LF6/DeisolationCapacity`, `Empirical/CSD/QEC/RegisterFlow` | | The flow of a `C¹` field on a compact manifold exists, obeys the group law and is jointly continuous (2026-09-11/12; Mathlib has no flows on manifolds) | `integralFlow_add`, `continuous_integralFlow` | `Mathlib/Geometry/Manifold/IntegralCurve/GlobalFlow`, `FlowContinuity` | Historical capstones (`FiniteQMClosure`, `unified_projectiveSector_capstone`, `measurement_flow_born_frequency`) stand unchanged; `unifiedArenaClosure` is their one-arena successor, with an explicit field-by-field mapping table in `RecordLayer/UnifiedArena.lean`. ## The story of measurement The dynamical arc is the corpus's most instructive chain, because every negative result became load-bearing: 1. **Constraints before construction.** `no_everywhere_correlation`: a continuous propagator cannot correlate everywhere, an exceptional set is forced (though not, as first over-read, discontinuity itself). `no_exact_collapse`: measure-preserving dynamics cannot contract; collapse must be *relocation with storage*. `collapse_accuracy_bound`: approximate collapse is priced in ready-state improbability, forcing Dirac calibration. 2. **The shear** (`ShearWitness`) creates Born-weighted, persistent records from a ready state, and provably *cannot* collapse (`shear_base_marginal_unchanged`). 3. **The calibrated swap** (`SwapWitness`, `SwapLuders`) adds the record-triggered bank exchange: collapse as relocation, rank-one Lüders as a pushforward. Its own boundary: `swap_not_blockLuders`, no fixed calibration serves degenerate blocks, because the demanded post-state depends on the preparation. 4. **The wall, diagnosed** (`BlockCollapse`): the degenerate mechanism exists at the vector level (`componentSwap`, collapse with the residual stored), but its ray-pair descent loses the *relative phase*. 5. **The join** (`PhaseSlot`, `JoinArena`): keep the phase, the pair arena *is* the projective join `ℙ(ℂ^{N+N})`, the swap becomes a permutation unitary, Liouville preservation becomes FS unitary invariance, and the update is pointwise deterministic (`join_block_luders`). 6. **The protocol and the obligation** (`JoinProtocol`, `JoinLuders`): the join update runs in the standard record architecture, and `joinWitness_blockLuders` inhabits the very obligation the no-go had closed off, from a fixed calibration. 7. **Covariance** (`RotatedSwap`): all of it, in every orthonormal basis (`measurement_covariance`). 8. **Classification** (`ShearDiscontinuity`, `PiecewiseHamiltonian`): what the exact-record witness's dynamics provably is, a piecewise rigid **symplectic** translation with null seam set (*not* globally Hamiltonian: the torus-flux correction of 2026-08-02). That raised the question the final step answers. 9. **The smooth horn** (`PointerArena` → `PointerGeneration`, eight modules): replace the torus register with a projective pointer `ℂℙ^N`, and the seam-jumping readout with a continuously modulated Hermitian coupling. The propagator becomes a ramped exponential, jointly continuous in time and state, Liouville-preserving, Schrödinger-generated (`rampedU_schrodinger`), that lands ready states in the record regions with the ontic sector selecting the outcome, at the price of a stated `ε` in records and Born (`smoothWitnessClosure`). The trade-off forced by `no_everywhere_correlation` is now held from **both ends**, each machine-checked: exact records with seams, or seamless dynamics with `ε`. *(Precision, 2026-08-03: the no-go rules out **everywhere**-exact records under continuity; the third option, continuous dynamics, records exact off a* **null** *seam, exact Born, was exhibited the same day (`nullSeamClosure`), at the price of the Dirac calibration. The fork is a trilemma; see §"Which horn is the right one?".)* ### Which horn is the right one? Neither, and that is a settled framing (author decision, 2026-08-03), not an open question. `no_everywhere_correlation` rules out **everywhere**-exact records for any continuous dynamics on a connected state space, and every measuring science has met that constraint. ### The three priced witnesses The constraint is sharp enough to be worth stating as a price list rather than a caveat. Records, exact Born and continuity are jointly impossible, so each formalised witness pays **exactly one** of three prices. Every one of the three is inhabited, which is what makes this a classification rather than an obstruction. | Witness | Price it pays | Where | |---|---|---| | Piecewise | **Seams**: records exact off a discontinuity set | `RecordLayer/` piecewise closure | | Smooth | **`ε`-Born**: continuous throughout, Born exact only to a stated `ε` | `smoothWitnessClosure` | | Null-seam | **Dirac calibration**: exact records *and* exact Born, at a calibrated ready point | `nullSeamClosure` (`RecordLayer/NullSeamWitness.lean`) | The finite-QM closure pays the first: its interaction is unitary and its Born weights are exact, so what gives is totality: `readout_ae_total`, records defined almost everywhere (`SigmaLayer/FiniteQMClosure.lean` names the horn). Two scope points, because the price list is easy to over-read. `no_everywhere_correlation` prices the **everywhere** case only: it does not bound the exceptional set's measure, so "a.e., with a null seam" is a coherent position and is the one the closure occupies. What forces a *positive-measure* no-record set is `posMeasure_noRecord_pointer` (`RecordLayer/NoRecordGeometry.lean`, 2026-08-05), on the pointer's own geometry, where continuity + open-map + positive-width ready + correlation leave no room. And **general exhaustiveness over all arenas is research, not a claim**. Three horns are exhibited and the fourth combination is refuted on one geometry; that is not a proof that no fourth horn exists anywhere. (*Precision added 2026-08-03, fourth external review; the candidate third option was **exhibited the same day**: `nullSeamClosure` (`RecordLayer/NullSeamWitness.lean`) is a continuous, measure-preserving witness with records exact off a **two-point** seam and **exact** Born, the transition crosses where the record regions kiss. Its price is the **Dirac calibration**: exactness holds at the calibrated ready point, and a positive-width ready region gives the no-record set positive measure (qualitative, the earlier "of order the calibration width" was withdrawn 2026-08-04; the order would need a landing-moment estimate). The honest classification is therefore a* **trilemma** *, each formalised horn pays exactly one of: seams (piecewise witness), ε-Born (smooth witness), Dirac calibration (null-seam witness), the last already priced by `collapse_accuracy_bound`. On the pointer's own geometry the fourth combination is now provably impossible, `posMeasure_noRecord_pointer` (`RecordLayer/NoRecordGeometry.lean`, 2026-08-05): continuity + open-map + open preconnected positive-width ready + correlation force a positive-measure no-record set. General exhaustiveness over all arenas remains research, not a claim.*) Digital electronics keeps continuous dynamics and engineers the `ε`: flip-flop *metastability* is an unresolved needle between the marks, its probability driven down exponentially with settling time, the classical twin of our no-go is Lamport's *Buridan's Principle* (1984), found by people building arbiter circuits. Thermodynamics keeps the exact jump and admits it is the infinite-size idealisation of a steep continuous crossover, bubble chambers and Geiger counters "jump" by phase transitions that are sharp only in the thermodynamic limit. Control engineering trades single-valued readout for hysteresis (Schmitt triggers); neurons near firing threshold do the same. Textbook QM alone promoted the jump to an *axiom*, collapse, instead of modelling the apparatus. The corpus's two witness families are these two universal answers, formalised: reach for the exact-record witnesses when the analysis needs sharp records (the operational closures do), and for the smooth witness when it needs honest Hamiltonian dynamics (the papers' architecture does). Neither is "the" CSD measurement; the trade-off theorem is. ## Three reading pathways, by reader **For the physicist**, what does CSD claim and what is actually proved? 1. [`specs/CSD-CHARTER.md`](../specs/CSD-CHARTER.md), the ontology and the anti-drift frame. 2. [`specs/reconstruction-status.md`](../specs/reconstruction-status.md) §2a, the A1–A7 audit. A2's **Hamiltonian-generation** sub-question, reopened 2026-08-02 on the torus-flux correction, was closed 2026-08-03 by the smooth pointer witness, at the formalisable level; §2a carries the full audit trail and the scoped residue. 3. `CsdLean4/RecordLayer/GlobalBasin.lean`, measurement partitions fixed by the apparatus. 4. `CsdLean4/RecordLayer/SwapLuders.lean` → `UnifiedArena.lean` → `JoinLuders.lean`, the measurement story above, in code. 5. [`EMPIRICAL.md`](../EMPIRICAL.md), every flagship experiment, QM proof + CSD ontic twin. **For the Lean/Mathlib reader**, how is it built and enforced? 1. [`CONVENTIONS.md`](../CONVENTIONS.md), the three-category discipline and naming rules. 2. [`AXIOMS.md`](../AXIOMS.md) §0, axioms vs. structural posits vs. foundational triple. 3. `CsdLean4/Tests/AxiomAudit.lean` + `scripts/check-claims.sh`, the enforcement: per-theorem `#print axioms` pins as `#guard_msgs`, and an epistemic-overclaim scanner over the docs. 4. `CsdLean4/Mathlib/` + [`MATHLIB-GAPS.md`](../MATHLIB-GAPS.md), the staged upstream candidates and the library gaps they answer. **For the skeptic**, where are the boundaries, and does the repo respect them? 1. [`AXIOMS.md`](../AXIOMS.md), start with what is assumed. 2. The five no-gos, in order: `no_everywhere_correlation`, `no_exact_collapse`, `shearEvolve_not_continuous`, `shear_base_marginal_unchanged`, `swap_not_blockLuders` (plus `SectorPostulateNoGo` for the sector-origin question). Each is load-bearing. 3. The ⚠️ **Honest scope** blocks in every module header, the corpus's standing convention. 4. [`specs/BACKLOG.md`](../specs/BACKLOG.md), the canonical open-items list, with effort grades. 5. The dated-corrections convention: errors are corrected *in place with dates*, never silently (see `specs/archive/` and the correction notes throughout the docs). ## Status and open work The A1–A7 reconstruction map's last genuinely open row, A2's Hamiltonian-generation sub-question, was closed 2026-08-03 at the formalisable level by the smooth pointer witness ([`specs/reconstruction-status.md`](../specs/reconstruction-status.md) §2a; the symplectic/moment-map *reading* of the generator remains the same scoped prose boundary as A1/A3). The dynamical measurement layer is complete through the capstone, rank-one, every basis, the degenerate package on one protocol, the smooth horn and its Schrödinger generation, mixed preparations *and their outcome-conditioned update*, dynamical no-signalling, the eraser process with its irreversibility theorem, and POVM/instrument dynamics via Naimark dilation. Three witness families now populate the measurement **trilemma** (seams, `ε`-Born, Dirac calibration), and the transition profiles have `C^∞` variants with the plateau interface unchanged. The two formerly-recorded extensions are now theorems: the `ε`-Born frequency layer (`pointer_born_frequency`, 2026-08-04) and the **smooth-witness Lüders composition** (`pointer_luders_marginal`, `RecordLayer/PointerLudersMarginal.lean`, 2026-08-05), conditioned on the outcome sector, the post-stroke system marginal is the slot calibration, so the smooth horn delivers records *and* a collapse-as-relocation update on one arena. The **local sharpened no-go is now a theorem** (`posMeasure_noRecord_pointer`, `RecordLayer/NoRecordGeometry.lean`, 2026-08-05): on the pointer's moment-region geometry, exact-a.e. records force Dirac calibration, the trilemma's third leg closed with no geometric hypothesis. General exhaustiveness over all arenas stays research (`specs/BACKLOG.md`). The empirical suite covers every flagship test on both branches. Connectivity claims are governed by [`specs/connectivity-manifest.md`](../specs/connectivity-manifest.md), nothing here may be read as stronger than a CONNECTED row there. Open work: [`specs/BACKLOG.md`](../specs/BACKLOG.md) (canonical), with the foundations frontier mapped in [`specs/sigma-fibre-contextuality.md`](../specs/sigma-fibre-contextuality.md).